Agentic Engineering

Engineers can use AI agents to handle most of the coding work. This skill gives you a clear system to manage those agents. You start by defining completion criteria before any code is written. Every task is broken into small, verifiable pieces that take about 15 minutes. Each piece has a single main risk and a clear done condition.

The skill uses an eval-first loop. You run capability tests and regression checks before and after each change. This catches failures early and shows what improved. Different AI models are used for different jobs. Simple tasks like classification use a fast model. Complex refactors use a smarter one. This keeps costs down while getting good results.

You also get rules for session management and code review. Focus reviews on security, edge cases, and hidden risks. Do not waste time on style issues that linters already handle. Track costs per task and only use expensive models when a cheaper one fails with a clear reasoning gap.